Ajent API

Ajent API is a modular, extensible Node.js API server for interacting with Large Language Models (LLMs) such as OpenAI (GPT-4) and Google Gemini (Vertex AI). It provides a unified HTTP interface for chat, streaming, and speech-to-text (STT) operations, with built-in support for retries, error handling, and custom hooks.

Features

• Unified API for multiple LLM providers (OpenAI, Vertex AI/Gemini, and more)
• Chat completions (single and streaming)
• Speech-to-text (audio transcription via OpenAI Whisper)
• Pluggable architecture: easily add new LLM providers
• Custom hooks for request/response/error handling
• File upload support for audio transcription
• Robust retry logic with exponential backoff and error simulation
• TypeScript-friendly (JSDoc types)
• Easy to embed or run standalone

Quick Start

1. Install dependencies

npm install ajent-api

2. Set up environment variables

For OpenAI:

• AJENT_LLM_TOKEN — Your OpenAI API key

For Vertex AI (Gemini):

• GOOGLE_APPLICATION_CREDENTIALS — Path to your Google Cloud service account JSON
• AJENT_LLM_PROJECT — Your Google Cloud project ID

3. Start the server

You can create a simple server using the provided API:

const { createLLMServer } = require('ajent-api');

const server = createLLMServer({
  llmName: process.env.LLM_NAME || 'openai', // or 'gemini'
  llmToken: process.env.LLM_TOKEN,
  llmProject: process.env.LLM_PROJECT, // for Vertex AI
  llmLocation: process.env.LLM_LOCATION, // for Vertex AI
  llmModel: process.env.LLM_MODEL || 'gpt-4.1' // or Gemini model name
  port: process.env.PORT || 3000
});

server.start();

Or run your own entrypoint that uses the API server.

4. API Endpoints

All endpoints are mounted under /agent:

POST /agent/message

• Body: { messages: [...], tools?: [...] }
• Response: { message: ... }

POST /agent/message/stream

• Body: { messages: [...], tools?: [...] }
• Response: [Server-Sent Events (SSE) stream]

POST /agent/audio_message

• Form-data: audio (file upload)
• Response: { transcription: ... }

5. Customization

You can pass hooks to customize request/response/error handling:

const server = createLlmServer({
  beforeRequest: (req, res) => { /* ... */ },
  afterResponse: (req, response) => { /* ... */ },
  errorHandler: (err, req, res) => { /* ... */ }
});

Project Structure

• src/api/factory-api-server.js — Main API server factory
• src/llm/llm-client.js — Abstract LLM client (retry logic, error handling)
• src/openai/openai-client.js — OpenAI implementation
• src/gemini/vertexai-client.js — Vertex AI (Gemini) implementation
• src/llm/response-serializer.js — Response serialization utilities
• src/utils/ — Utility modules (logger, converters, etc.)

Testing

Run tests with:

npm test

License

MIT

Ajent API — Unified, extensible LLM API server for modern AI applications.